1. Field of the Invention
The present invention relates to a method of online detecting MP3 music transmissions and an architecture thereof, particularly to a method of online detecting MP3 music transmissions and an architecture thereof, which utilizes an MP3 packet capture technology and an MP3 content-based retrieval technology to detect illegal MP3 transmissions.
2. Description of the Related Art
Owing to the explosive advance of the hardware and the extensive prevalence of the Internet, the digital music has been popularized globally. MP3 (MPEG 1 Audio, Layer 3) is a compression format of high-quality and high-compression ratio and dominates in audio-file storage and transmission. However, it is exactly the high quality and high compression ratio that makes illegal MP3 music files flood all over the Internet, which infringes upon the intellectual property seriously and brings about the loss obviously, and the authorities have also suffered from the illegal behavior for long because of lacking an effective inhibition mechanism.
Just like that the fingerprint can identify a person, the MP3 fingerprint can identify an MP3 file. Cano, et al. pointed out: The audio fingerprint is usually constructed with a digital signature or a multi-dimensional feature value, and some methods, such as the Euclidean method, are used to calculate the similarity thereof; in order to increase the comparing efficiency, some data-structure technologies are used to filter out less-possible samples and reduce the work-load of comparing calculation. However, extra-added data, such a digital signature, may be doctored; besides, most MP3 files are not appendixed with a digital signature. Therefore, the digital signature does not benefit the identification of MP3 files very much. Contrarily, the feature-value fingerprint is derived from the file itself. Even though the outward data is incorrect or modified, the real identity can still be acquired from the file itself. At present, the related researches can be roughly grouped into the fingerprint-construction part and the fingerprint-identification part.
The current MP3-identification technology and related literatures is stated blow.
Haitsma, Kalker, Oostveen et al. derive a 32-bit sub-fingerprint from per 11.6 msec of data, and 256 derived sub-fingerprints are integrated into about three seconds of fingerprint of a song. When the comparison is performed, each of the 256 sub-fingerprints is compared individually, and the similarity is determined by the integrated result of the 256 comparing calculations. As the comparing result depends on the 256 separate comparing calculations, the fault tolerance of the comparing calculations increases.
Lu used two kinds of fingerprints—the authentication fingerprint and the identification fingerprint—to perform identification. Firstly, the authentication fingerprint is used to find out possible samples, and then, the identification fingerprint is used to perform detailed comparing operation; thereby, the calculation cost is effectively saved.
Burges, Platt, Jana, et al. proposed a Distortion Discriminant Analysis method to extract out the feature values, wherein the Oriented Principal Components Analysis process is repeated twice to concentrate the 2048-dimensional audio information into a 64-dimensional coefficient in order to relieve the high-dimension problem. Besides, the imperceptible portion of the audio information is also removed to reduce the data quantity.
Sukittanon and Atlas proposed a feature value combining audition and frequency-modulation, wherein 15 seconds of audio data is firstly processed by Fourier transformation and then decomposed by frequency-modulation and lastly subband-normalized in order to reduce time and frequency distortions.
Herre, Hellmuth, Cremer, et al. pointed out that just MPEG-7 itself defines two structures—the low level descriptor and the description scheme, which may be regarded as fingerprints and can be used to describe audio information; the low level descriptor can be used to extract out a feature value, and the description scheme can be used to classify feature values, and thereby, the contents of audio information can be correctly described.
Seifer proposed an LSM (Leadsheet model), wherein four steps—pitch estimation, pitch aggregation, hypothesis creation, and hypothesis confirmation—are used to describe the melody, rhythm, accompaniment, etc. of a song.
Miller, Rodriguez, Cox et al. embedded a given fingerprint into a 256-dimensional hierarchical structure. Via the correlation format constructed with the hierarchical structure, the depth of the hierarchical structure and the bit error rate, the closest answer can be obtained.
It is ineffective and laborious to on-side search a suspected illegal transmission; therefore, the present invention proposes a method of online detecting MP3 music transmissions and an architecture thereof, which effectively detects and records the IP addresses of the sender and the receiver of an illegal MP3 music transmission in order to implement investigation and secure musical intellectual property.